SuperLender is a digital lending company struggling with loan defaults. Using customer demographics, loan history, and repayment behavior, this project builds machine learning models to predict the likelihood that a new loan will be repaid (“Good”) or default (“Bad”). The analysis revealed that loan amount, repayment history, and employment status were among the strongest predictors of creditworthiness. With the best-performing model (Random Forest, 95.87% accuracy), SuperLender can automate risk scoring and reduce default rates by focusing on high-risk applicants before disbursing loans.

1. traindemographics.csv – Customer demographic information
2. trainperf.csv – Performance of current loans (target: good_bad_flag)
3. trainprevloans.csv – History of previous loans
4. test... files – Similar structure for testing

• Data cleaning (handling nulls, datetime conversion)
• Feature engineering (loan ratios, repayment gaps, etc.)
• Correlation analysis via heatmap
• Multiple ML models:
  • Logistic Regression
  • SVM
  • KNN
  • LightGBM
  • XGBoost
  • Gradient Boosting
• Model comparison and evaluation using accuracy, ROC-AUC, and confusion matrix.

-Converted date columns to datetime Removed null values and irrelevant columns Merged all train and test data into unified datasets

• Feature Engineering Created ratios such as totaldue / loanamount Calculated repayment duration and loan term features Encoded categorical variables (education, bank name, etc.)
• Exploratory Data Analysis Correlation analysis with heatmap Visualization of loan amount distribution and term duration Identification of key predictors for default.

-Customers with shorter repayment gaps and stable employment have better repayment behavior. -Loan amount and education level strongly influence credit performance. -Ensemble models (Random Forest, LightGBM, XGBoost) achieved the best overall accuracy and ROC-AUC.

• LightGBM achieved the best performance (highest ROC-AUC).
• Feature importance plots highlight key predictors of loan repayment.

• Python 3
• Pandas, NumPy
• Scikit-learn
• LightGBM, XGBoost
• Matplotlib, Seaborn
• Jupyter Notebook